[cfe-dev] Startling optimization

[Xi Wang]

you may try the following passes that will optimize the loop away.

clang -emit-llvm -S -o - test.c | opt -mem2reg -loop-rotate -instcombine -indvars -loop-deletion -S

- xi

On Jun 20, 2011, at 12:56 AM, Robert Purves wrote:

> $ cat test.c
> int foo( int n ) 
> { 
>  int  a = 0, b = 0, c = 0, j; 
> 
>  for ( j = 0; j < n; j++ ) 
>  { 
>    a++; 
>    b = a | c; 
>    c = a; 
>  } 
>  return b; 
> }  
> 
> $ clang -O -S test.c
> 
> 
> _foo:
> 	pushq	%rbp
> 	movq	%rsp, %rbp
> 	testl	%edi, %edi
> 	jle	LBB0_2
> 	leal	-1(%rdi), %eax
> 	orl	%edi, %eax
> 	popq	%rbp
> 	ret
> LBB0_2:
> 	xorl	%eax, %eax
> 	popq	%rbp
> 	ret
> 
> Which parts of clang/llvm source should I read to learn how the loop is magically optimized away?
> 
> Robert P.
> 
> 
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev